-
-
Meet GhostRep — AI-assisted training with a second perspective when you need it.
-
A dedicated coach inbox — Access approved training context in a separate example workflow.
-
Training that fits you — Build around your goals, equipment, experience, and restrictions.
-
Review before delivery — Coach feedback requires explicit approval before reaching the athlete.
-
Approve with confidence — Review the price and separately approve sharing and purchase.
-
Pay through PayPal — Approve checkout securely using PayPal’s Sandbox environment.
-
Find your coaching fit — Share your preferences and request a session time.
-
Clear payment status — Pending payment keeps coaching fulfillment locked until confirmation.
-
Your plan, your next step — Review your training week or ask a live coach.
-
Visibility for the seller — View the matching transaction and its status in PayPal.
-
Context for better coaching — Review the shared plan, preferences, and requested appointment.
-
Choose informed support — Compare services, transparent prices, and evidence-based recommendations.
-
Close the feedback loop — Approved coaching notes appear on the athlete’s receipt in the example workflow.
-
See the evidence — Replay specific repetitions and explore movement observations.
-
Capture your movement — Record or upload a barbell squat set for analysis.
Inspiration
An AI-generated workout can tell you what to do—but it does not always give you the confidence to do it.
Does this plan fit my restrictions? What changed in that repetition? Should I ask a coach before continuing?
GhostRep grew from that gap between receiving an answer and deciding what to do next. It connects personalized training, AI-assisted movement analysis, and optional expert support in one workflow.
For the PayPal hackathon, we extended the existing training application with a commerce layer that turns a training question into an informed service recommendation, an athlete-approved purchase, and a trackable coaching request.
What it does
GhostRep helps athletes plan, understand, and choose their next step.
Athletes start with their goals, equipment, experience, schedule, and training restrictions. They review a generated training week before activating it, with an option to request a live coach’s perspective directly from the plan.
For supported barbell squat recordings, GhostRep uses AI pose estimation to track body landmarks and compare repetitions against a reference from the athlete’s own set. Observations link to specific repetitions, so athletes can replay the evidence behind the feedback. The reference is a personal comparison point—not a claim of perfect technique.
When additional support could help, the commerce agent recommends from a fixed catalog:
- AI Technique Review — $4.99
- Expert Video Review — $14.99
- Personal Coaching Session — $29.99
The agent explains its recommendation, offers alternatives, and can account for a budget. It does not invent prices or purchase automatically.
For a live coaching request, athletes can select preferences such as language, gender preference, coaching experience, and focus, then suggest a session time. Matching and availability require coach confirmation.
Sharing training context and approving a purchase are separate decisions. PayPal handles checkout, and GhostRep verifies payment on the server before unlocking the purchased service.
A separate authenticated coach workspace supports reviewing shared context, managing appointment requests, and preparing feedback. Feedback stays private until the coach explicitly approves delivery to the athlete’s receipt.
How we built it
We built the frontend with Next.js, React, TypeScript, and Tailwind CSS, and the backend with FastAPI, Python, SQLAlchemy, and SQLite.
MediaPipe pose estimation provides the AI foundation for movement tracking. OpenCV and FFmpeg support video processing and replay. GhostRep turns tracked landmarks into repetition comparisons and evidence-linked observations.
An optional language-model coaching layer can explain saved evidence conversationally. Source labels distinguish measurements, model interpretation, and local summaries.
Commerce recommendations use deterministic matching rules against a server-owned catalog. This keeps prices and purchase behavior predictable: model-generated text cannot authorize a transaction or override payment verification.
The PayPal integration uses Orders v2 and the JavaScript SDK v6, with server-side capture, reconciliation, refund handling, and webhook signature-verification support.
The complete workflow connects:
Training context → recommendation → sharing consent → purchase approval → PayPal checkout → verified payment → fulfillment
The existing training and analysis foundation predates this hackathon edition. The PayPal version adds the consented commerce workflow, payment controls, plan-to-coach handoff, preference snapshots, and authenticated fulfillment tools.
Challenges we ran into
Keeping training context intact through checkout.
A payment button alone would not solve the athlete’s problem. We needed to carry the selected service, question, approved training context, and coach preferences into a request that a coach could act on.
Handling payment uncertainty correctly.
Buyer approval, completed payment, and appointment confirmation are separate events. Our recorded PayPal Sandbox transaction entered pending review. GhostRep preserves that state and keeps fulfillment locked rather than treating approval as completed payment or asking the athlete to pay again.
Separating evidence from interpretation.
A movement difference does not automatically establish whether an exercise is suitable or technically correct. We linked observations to replay and kept measured information distinct from coaching interpretation.
Building the service after the sale.
Checkout needed a meaningful destination. That required coach authentication, shared-plan review, appointment handling, feedback drafts, explicit delivery approval, and an athlete-facing receipt.
Accomplishments that we're proud of
We connected a specific training question to an optional paid service without losing the athlete’s intent or supporting evidence along the way.
We’re particularly proud of:
- Making live coach review accessible directly from a generated plan.
- Connecting movement observations to replayable repetitions.
- Giving recommendations clear reasoning, fixed prices, and alternatives.
- Separating data-sharing consent from purchase approval.
- Keeping service access locked until payment is verified.
- Building both the athlete experience and the coach delivery workflow.
- Preserving the existing planning, training, and analysis flows while adding commerce.
We also prepared a clean, runnable source package with setup instructions, dependency lockfiles, and automated tests. Backend and frontend checks, lint, and the production build passed.
What we learned
Agentic commerce becomes useful when the handoff between assistance and action is clear.
The agent should explain why a service might help. The athlete should control what is shared and purchased. Payment verification should determine access. The coach should receive enough approved context to deliver a useful response.
We also learned that pending states deserve as much attention as successful ones. A trustworthy application should explain what happened, what remains unresolved, and what the customer should do next.
Finally, AI does not need to control every decision. Combining AI-based movement tracking and optional conversational interpretation with explicit commerce rules made the workflow easier to understand, test, and trust.
What's next for GhostRep
Our next priorities are to:
- Validate a completed PayPal Sandbox capture and verified webhook delivery end to end.
- Provision coaches for real service delivery and improve availability matching.
- Expand movement analysis beyond barbell squats.
- Evaluate coaching explanations against saved evidence.
- Improve the feedback loop between coach recommendations and future training.
The demonstration shows an actual PayPal Sandbox transaction pending review. Coach delivery is demonstrated using a clearly identified separate example order with simulated payment, rather than presenting it as fulfillment of the pending transaction.
Our goal is to make GhostRep a practical bridge from AI insight to useful action—with the athlete in control of every purchase.
Built With
- ai
- cv
- fastapi
- ffmpeg
- mediapipe
- next.js
- opencv
- paypal
- paypal-javascript-sdk
- paypal-orders-api
- python
- react
- sqlalchemy
- sqlite
- tailwind-css
- typescript